All terms
Hardware & Systems
Arithmetic Intensity
The ratio of compute operations to memory traffic, used to spot the bottleneck in a workload.
Definition
Arithmetic intensity is the ratio of floating-point operations performed to bytes moved to and from memory, usually given as FLOPs per byte. It indicates whether a computation is memory-bound (low intensity, limited by bandwidth) or compute-bound (high intensity, limited by raw throughput). Large batched matrix multiplications have high intensity, while single-token decoding has low intensity. The roofline model uses it to predict achievable performance.